Supply Chain Planning Considering the Production of

adfa, p. 1, 2011.

© Springer-Verlag Berlin Heidelberg 2011

Supply Chain Planning Considering the Production of

Defective Products

Ferrara Miguel, Corsano Gabriela, Montagna Marcelo

INGAR – Instituto de Desarrollo y Diseño CONICET-UTN

Avellaneda 3657, Santa Fe, Argentina

E-mail: {miguelferrara, gcorsano, mmontagna}@santafe-

conicet.gov.ar

Abstract. When the markets become more competitive, the customer satisfac-

tion level starts to be a more important factor. One of the principal components

to achieve the customer satisfaction is to minimize the amount of defective or

non-standard products. Therefore, it is convenient to consider decisions about

product quality when the supply chain is planned. In this work, a mixed integer

linear programming mathematical (MILP) model is developed for the supply

chain planning, which determines impact of different decisions over the ex-

pected number of defective products. These decisions are: supplier selection,

inbound control methods selection, and production process selection. The ob-

jective function is the cost minimization associated to the supply chain plan-

ning.

Keywords: Supply Chain; Planning; Customer Satisfaction; Defective Products

1 Introduction

Many companies when planning their supply chain (SC) face the challenge of taking

the right decisions between two conflictive objectives: Cost versus Costumer service

level. Without a satisfied customer, the supply chain strategy cannot be deemed effec-

tive. Van Hoeket al. [1] emphasized that to assess supply chain performance, supply

chain metrics must center on customer satisfaction. One of the principal ways to

achieve it is to deliver the least amount of defective products. In order to do that, a

model for the planning of the SC that considers the amount of defective products

delivered to each client is proposed in this work to assess the Customer service level.

There are few works in the literature considering the quality of the final products

delivered to the client for the design or planning of the SC [2, 3]. Some exceptions

are, for example, Franca et al. [4], who develop a multi-objective stochastic model for

3º Simposio Argentino de Informatica Industrial, SII 2014

43 JAIIO - SII 2014 - ISSN: 2313-9102 - Página 209

this problem. The quality of the total production is an objective function, and it is

measured considering the quality of raw materials used for producing each product.

Chung-Chi et al. [5] consider one supplier and one manufacturer both with imperfect

production and inspection systems. The model establishes the amount of investment

to increase the quality of the production and inspection system in order to maximize

the expected profit per unit. Brojeswar et al. [6], propose a continuous- time produc-

tion and inventory model considering defective raw materials and rework of defective

products. The manufacturer engages in 100% perfect inspection and discards the de-

fective raw materials, so they do not enter in the production process. Duffuaa et al. [7]

develop a multi-objective optimization model for a process targeting problem. The

quality of the product is a random variable with known variance, and the objective is

to find the optimal value for its mean in order to maximize the expected profit per

product, along with other objectives. A product is considered non-conforming if its

quality is lower than a given value, and, in this case, it is sold at lower price to the

client.

2 Problem Description

A MILP model is presented to solve a SC planning problem of one period. The SC

considered in this work involves three echelons: suppliers, production plants, and

clients. The product demands of each client must be fulfilled respecting the maximum

amount of defective products that each client tolerates. The impact of 3 types of deci-

sions on the product quality is considered: suppliers selection, selection of in-bound

control methods, and selection of production processes.

The model determines the following planning decisions: how much raw materials

must be purchased at each supplier and which inbound control systems are used. Also,

how many products are produced by each production process (technologies) at each

plant, how many defected products are discarded, and how many defective and non-

defective products are sent to each client.

It is assumed that a type of raw material to produce a given product, at a given plant,

must be purchased at a single supplier, and controlled by a single inbound control

method. It is also considered that all the units of raw material of a certain type, deliv-

ered from a certain supplier, have the same probability to be defective. This probabil-

ity is usually related with the price, in the sense that it will be higher if the price is

cheaper. The raw materials that arrive to the plants are controlled by inbound control

methods that are imperfect; this means that there is a probability of not detecting de-

fective raw materials. This implies that, defective units of raw material may enter into

the production system, causing a probability of producing defective products. The

units of raw material detected as defective are automatically discarded and they do not

enter in the production process. The best control method will be more expensive. The

raw materials not discarded are used in any of the available production process in the

plant. At each plant the model decides how many units are produced by each produc-

tion process (technologies). A unit of product produced by a given technology has a

known probability to be defective.



Therefore, according to the supplied raw material, inbound control method and

process technology, a unit of final product will have a probability to be defective that

depends on which combination of these decisions has been taken to produce it. Then,

the expected number of produced defective products can be computed. Also, there is a

penalty for each unit of defective final product delivered to a client. This penalty de-

pends on the client. At each plant, 100% perfect outbound inspection over its final

products is performed. This means that the total amount of defective products will be

detected. After this inspection the model decides how many defective products are

discarded and how many are sent to each client. This decision is made based on the

tradeoff between the cost to produce a unit of product versus the penalty cost of deliv-

er a defective unit of product. The objective function considers the sum of the follow-

ing costs: raw materials, control, production, transport, and penalty. Finally, the

model determines the SC planning with minimum cost that satisfies the demands of

the clients and their tolerance for defective products.

3 Mathematical Model

In this section, the mathematical formulation is presented. The model parameters and

variables are defined in the Nomenclature section.

Selection of Suppliers and Inbound Control Systems for Raw Materials.

The Equations 1 and 2 determine which plants are going to be used and which prod-

ucts they are going to produce.

(1)

(2)

Let = be the set of raw materials necessary for producing

product i, and , the set of all raw materials. Let denote an element of

simply by r. It is assumed that only one supplier can supply raw material r for the

production of i in plant l if that product is produced in l:

(3)

For a plant l, an inbound control method u with 100% inspection is used to control the

raw material r to produce i only if the product i is produced at plant l:

(4)

To make zero if is zero, a constraint of big-M type is stated:

(5)



Defective Raw Materials Detected by the Control System

According to the concepts of random variable and Binomial distribution

denotes that a random variable has Binomial distribution with parameters

(the number of trails) and (the probability of success). In order to denote the ex-

pected value of a random variable the notation is used.

It is assumed that suppliers can deliver defective raw material. Supposing that a sup-

plier delivers N units of raw material r, let be a random variable that measures the

quantity of defective units of raw material r delivered by s. We suppose

that where is a parameter that represents the probability for a

unit of raw material r supplied by supplier s to be defective

On the other hand, each inbound control system u is imperfect. This means that a

defective raw material r can be undetected by u, but a non-defective raw material r

will never be detected as defective. The units of raw material are controlled one at a

time by the control system. Given a unit of defective raw material r, the probability

that the inbound control system u detects it as defective is the parameter . Also,

when a unit of raw material r is detected as defective then it is automatically discard-

ed. Then

As this equation must be satisfied only when raw material r comes from s for produc-

ing i in l and it is controlled by u, we can state this using constraints of Big-M type

(6)

(7)

,

Now, the raw material that is not discarded is used to produce. This is stated in the

following two equations:

3.3 Probability for Non-discarded Unit of Raw Material to Be Defective

In order to calculate the probability for a unit to be defective given that it has not been

discarded, i.e. defective raw material not detected by the control system, two well

known results of probability theory has been used: the Bayes theorem and the law of

total probability.

=



In other words, the probability that a unit of raw material r from s controlled by u

remains defective after the discarding is equal to the parameter . Let be

random variable that measures the quantity of raw material r from s that is defective

but not detected by u and it is used to produce a unit of product i in l. Then, because

of previous suppositions; . Where is a parameter that de-

notes the amount of raw material r necessary to produce a unit of product i. This pa-

rameter must be a natural number.

3.4 Probability to Produce a Defective Unit of Product.

Defective Products Due to Defective Raw Materials

In the model we consider that one of the reasons for a product to be defective is to

produce it with an amount of defective raw material that exceeds a threshold.

The parameter , with is defined as follows: if a unit of product i is

produced with more than units of defective raw material , then the unit of

product i is considered defective, otherwise it is considered non-defective. If at least

one unit of defective raw material will produce a defective unit of product i then

is set to zero. If the condition of raw material does not affect the quality of

the product i then , this means that every unit of used to produce i

can be defective.

The probability for a unit of product i produced in plant l to be defective due to de-

fects in the raw materials must be determined. Then, the selected suppliers for each

raw material to produce product i in plant l must be known, as well as, i.e. the pro-

curement policy-, the chosen control system to check each raw material, i.e which

was the inbound raw material control policy.

Let be a list of elements of , with

is the supplier of raw material to produce i in l, is a raw material pro-

curement policy for product i in plant l. Let be the set of all raw material procure-

ment policies for product i in plant l. Let be a list of elements of

with , where is the inbound control method ap-

plied to raw material to produce product i in plant l, is an inbound raw

material control policy for product i in plant l. Let be the set of all inbound raw

material control policies for product i in plant l.

Now we compute the probability for a unit of product i produced in plant l to be de-

fective due to defects in the raw materials.

Let be the event “a unit of product i produced in plant l is defective due to the

use of defective raw materials from the set , where is the raw material procure-

ment policy and is the inbound raw material control method policy”. In order to

calculate the random variable , defined in section 3.2, is used:



From these equations, it can be noted that is a function of i, l, and ,

Let define the parameter

Defective Products Due to Defects of the Technology

A unit of product i can be defective not only because of defective raw materials but

also because of defects of the productive process or technology t. Let be the event

“a unit of product i produced in plant l by technology t is defective due to defects in

the technology t”. Suppose that where is a parameter. Let be

a random variable that measures the quantity of defective products i produced by

technology t due to defects of the technology t, given that N units of product i are

produced by technology t. We suppose that

Defective Products Due to Defective Raw Materials or Defects of the

Technology

Let be the event “one unit of product i produced in plant l, is defective due to

any or both of the next reasons: “

The use of defective raw materials from the set where is the raw material

procurement policy and is the inbound raw material control method policy

Defects in the technology t.

Then

But . Let define the parameter with

. It measures how much increases when defective raw materials are

used to produce product i using technology t. Then for simplicity we assumed that

and so the parameter .

Let’s define the parameter

3.5 Calculation of the Expected Amount of Defective Units of Product

Let be a random variable that measure the quantity of defective product i due

to the event when units of product i are produced by t in l with purchase and

control policies and , respectively. It is assumed that .

Let be the expected number of defective products i produced in plant l using

technology t due to the event , given that the purchase and control policy were

and respectively. So

Suppose that is a feasible solution of the problem. In order to compute for

this feasible solution, suppose that has a purchase policy

and an inspection policy . For :

(



So this condition is stated using a big-M restriction:

with and (10)

with and (11)

Restriction (10) and (11) are active only when

3.6 Fulfillment of the Demand

A 100% outbound inspection is assumed to be applied at each plant, i.e. with a control

method that is perfect. So the expected number of defective products detected is equal

to the expected number of defective products.

(12)

(13)

(14)

(15)

(16)

Equation (12) is a mass balance between the defective products produced, the defec-

tive products sent to each client, and the defective products discarded. Equations (13)

to (16) state the mass balance of non-defective products to clients, the compliance of

the tolerance of defective products of each client, and the fulfillment of the demand.

3.7 Objective Function

The objective function is to minimize the total cost:

(17)

is a positive variable, it represents the inbound control method costs. The

constraints to settle its value are not shown for space reasons, but they can be derived

from the model presented here.



4. Examples

Case Study 1. Consider a SC with 3 suppliers (s1-s3) which can provide 3 types of

raw material (r1-r3) to 2 production plants (l1 and l2) where 3 products (i1-i3) can be

produced to cover the demand of 2 clients (k1 and k2). Also 2 inbound control meth-

ods (u1 and u2) can be used at any plant to control any raw material, and 2 technolo-

gies (t1 and t2) can be used at any plant to produce any product. In Table 1, 2 and 3

the model parameters are shown. Parameters can be obtained from the values

of the parameters in Table 1, 2 and 3. The rest of the parameters are not shown for

space reasons, but they are available for the interested reader. The MILP model was

implemented and solved in GAMS with an Intel (R) Core2Duo, 2.66 GHz using the

solver CPLEX. It has 7920 equations, 1966 continuous variables, and 98 binary vari-

ables. A solution with a 1% optimality gap was found in 550 CPU s.

Table 1. Parameters for Case Study 1

Raw materials s1 s2 s3 s1 s2 s3 u1 u2 u1 u2

r1 0.165 0.15 0.108 0.1 0.13 0.15 0.001 0.0015 0.9 0.95 r2 0.16 0.2 0.18 0.12 0.08 0.09 0.001 0.0013 0.87 0.93

r3 0.15 0.13 0.2 0.12 0.15 0.1 0.001 0.0013 0.9 0.96


Products

r1 r2 r3 r1 r2 r3 t1 t2 t1 t2 k1 k2 k1 k2

i1 3 4 6 0 0 0 0.4 0.6 0.07 0.03 0.12 0.12 3.5 7.5

i2 4 5 5 0 2 0 0.5 0.8 0.06 0.05 0.12 0.12 3 5 i3 2 3 3 0 0 0 3 3.2 0.05 0.03 0.12 0.12 3.5 4


Pro-ducts

t1 t2 k1 k2

i1 1.11 1.09 1500 1300 i2 1.08 1.08 1000 1200

i3 1.04 1.07 900 1100

Figure 1 shows SC flows for the solution. The total cost was equal to $ 27,369. Table

4 shows that the solution includes different decisions for each product regarding how

much percentage of defective products are delivered. Once a product has a given pro-

curement policy, control policy, and it is known how many units are going to be pro-

duce with each technology, the model determines the percentage of defective products

delivered to each client. The main factor to make this decision is the relation between

the parameter and the sum of all the necessary costs to produce a unit of that

product. For example, for product i3 the percentage of defective products delivered is

at its maximum for each client, because the sum of all the necessary costs to produce

a unit of product i3 is $4.26 which is higher than the penalty costs for both clients

.So, in this case, it is more convenient to deliver a defective



unit of product and to pay the penalty. Conversely, for product i1, there are 2915 units

produced by t1 and 725 produced by t2. The cost to produce a unit with t1 is $2.4, and

$2.6 with t2. Because these costs are lower than the penalty cost for both clients

, it is more convenient to produce again. Figure 1 shows

that the decisions about procurement, control policy, and the technology used to pro-

duce are almost the same for all the products. It can be noted that u2 is always chosen

except to control r2 for product i2. This is because of so is less likely

that defective raw material r2 can provoke a defective unit of i2. Therefore, it is not

important to use a more accurate control system. The procurement policy is the same

for all products which consist of the cheapest and worst suppliers in terms of quality.

This is based on the fact that control systems are so accurate and cheap in relation

with the purchase cost of the other suppliers, that is better to buy and discard more

raw materials, than to pay a higher price for raw materials and decrease the discarded

materials. Also the defected materials are detected with a very high probability so

their impact on the production is not important. In Figure 1 the notation s.r.i.u=x rep-

resents that supplier s supplies x units of raw material r to produce i and are controlled

by u. The notation i.nd=x represents that x non-defective units of product i are sent to

a client and i.d=x represents that x defective units of product i are sent to a client.

Fig. 1. SC flows for Case Study 1 solution

Table 4. Information of the solution of Case Study 1

Products Percentage of defective products

delivered (%)

Amount pro-

duced

Amount dis-

carded

Penalty cost

paid k1 k2

i1 0 0 3340 540 0

i2 12 0 2348 184 360

i3 12 12 2000 0 863

Case Study 2. In this example a variation on the values of the penalties is performed

in order to analyze the changes that it causes on the solution of the study case 1. Table

5 shows the new penalties in the last column.

s2

l1

k1

k2

s3

s1.r2.i1.u2=15040

s1.r2.i2.u1=13314 s1.r2.i3.u2=6753

s1

s2.r3.i1.u2=23410

s2.r3.i2.u2=13930

s2. r3.i3.u2=7009

s3.r1.i1.u2=11680

s3.r1.i2.u2=11124

s3.r1.i3.u2=4664

i1.nd=1500;

i2.nd =880; i3.nd=792;

i2.d =120; i3.d=108

i1.nd=1300;i2.nd=1200

i3.nd=978 ; i3.d=121



Table 5. Penalties and information of the solution of Case Study 2

Products Percentage of defective

products delivered (%)

Amount

produced

Amount dis-

carded

Penalty

cost paid

k1 k2 k1 k2

i1 12 12 2921 121 695 2 2.15 i2 12 0 2372 172 360 3 5

i3 0 0 2250 250 0 6 7

A solution with 1% optimality gap for this new case was found in 1900 CPU s. The

total cost was equal to $26,301. In this solution the procurement and control policies

for raw materials are the same as in study case 1 and also only the plant l1 is used to

produce. The most significant changes are in the percentage of defective products

delivered to each client, and in the amount of produced and discarded products. Table

5 shows these values that confirm some of the conclusions for study case 1.

5. Conclusion

In this work a model is presented to integrate decisions that impact on the quality of

the produced product with decisions regarding the planning of the SC. This approach

provides a useful tool for the evaluation of the quality impact on different scenarios of

provider selection, inbound control methods, production processes, and percentage of

defective products delivered and produced, and also to assess the trade-offs between

all of them. In order to add uncertainty to the model, some decisions are represented

using a probability distribution. The problem was formulated as a MILP model, where

several variables represent expected quantities. This model is a tool to support deci-

sion making in order to face negotiation processes with suppliers and clients. Also,

the knowledge of the impact of the quality of the raw materials over the quality of the

production is useful to establish more fruitful and cooperative relation with them.

It is worth highlighting the importance of combining the decisions about SC planning

with the ones that affect the product quality in the competitive context of modern

world. Numerical examples were presented and solved with the proposed approach.

Case Study 2 shows how the changes in the penalties from the clients affect planning

decisions regarding production.

Nomenclature

Indices:

Sets:



Binary Variables:

and 0 otherwise

Positive Variables:

Parameters:



References

1. van Hoek, R.I. “The contribution of performance measurement to the expansion of

third party logistics alliances in the supply chain”. International Journal of Opera-

tions & Production Management 21, 15–29, (2001)

2. Farahani R.Z., Rezapour S., Drezner T., Fallah S. “Competitive supply chain net-

work design: An overview of classifications, models, solution techniques and ap-

plications”, Omega, in press, accepted in (2013).

3. Mula J., Peidro D., Díaz-Madroñero M., Vicens E. “Mathematical programming

models for supply chain production and transport planning”, European Journal of

Operational Research, 204, 377–390, (2010)

4. Franca R. B., Jones E.C., Richards C.N., Carlson J.P. “Multi-objective stochastic

supply chain modeling to evaluate tradeoffs”, Int. J. Production Economics, 127,

292–299, (2010)

5. Hsieh C.C., Liu Y.T., “Quality investment and inspection policy in a supplier–

manufacturer supply chain”, Euro. J. of Oper. Research, 202, 717–729, (2010)

6. Pal B., Sana S.S, Chaudhuri K. “Three-layer supply chain – A production-

inventory model for reworkable items”, Applied Mathematics and Computation,

219, 530–543, (2012)

7. Duffuaa SO., El-Ga'aly A., “A multi-objective optimization model for process tar-

geting using sampling plans”, Computers & Industrial Engineering, 64, 309–317,

(2013)



Documents

Supply Chain Planning Considering the Production of